Octagram Propagation Matching for Multi-Scale View Stereopsis (OPM-MVS)
Multiple view stereopsis (MVS) algorithms have achieved substantial advancements in 3D reconstruction accuracy and completeness. However, significant challenges persist with texture-less regions, occlusions, thin structures, repetitive structures, and non-Lambertian surfaces due to unreliable photom...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11003953/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1849328753622122496 |
|---|---|
| author | Ray L. Khuboni Hongjun Xu |
| author_facet | Ray L. Khuboni Hongjun Xu |
| author_sort | Ray L. Khuboni |
| collection | DOAJ |
| description | Multiple view stereopsis (MVS) algorithms have achieved substantial advancements in 3D reconstruction accuracy and completeness. However, significant challenges persist with texture-less regions, occlusions, thin structures, repetitive structures, and non-Lambertian surfaces due to unreliable photometric consistency and insufficient geometric information. To address these challenges, we propose octagram propagation matching for multi-scale view stereopsis (OPM-MVS), a novel framework for efficient and accurate depth map estimation. The framework introduces octagram checkerboard propagation as a structural sampling method that enhances hypotheses propagation by aggregating cost information from overlapping neighbouring regions along structural lines. This approach ensures accurate and consistent hypotheses selection while improving spatial propagation efficiency. To refine depth estimation, we incorporate multi-scale depth confidence-guided geometric consistency, propagating reliable depth estimates from coarse to finer scales and minimizing propagation error. A dual bilateral weighted normalized cross-correlation is incorporated to reduce irrelevant pixel influence and improve correspondence in texture-less and repetitive regions. A multi-hypotheses joint view selection strategy is guided by neighbouring joint view probability to aggregate the subset views for cost matching and integrates effectively with the proposed propagation method. Experimental evaluations on public benchmarks includes ETH3D and Tanks and Temples to demonstrate the effectiveness of OPM-MVS in achieving superior completeness while maintaining comparable accuracy to COLMAP, ACMH, ACMM and other surveyed MVSNet methods. The framework delivers state-of-the-art performance by accurately reconstructing depth in challenging regions resulting in complete and reliable 3D models. Notably, OPM-MVS achieves an F1-score of 83.50% on the overall ETH3D test dataset under a 2 cm threshold which outperforms COLMAP (73.01%), ACMH (75.89%), ACMM (80.78%), DP-MVS (83.11%), MVP-Stereo (76.12%), TransMVSFormer (75.39%), TAPA-MVS (79.15%), and UGNet (80.83%). These results establish OPM-MVS as a promising solution for high-resolution and large-scale multi-view stereo applications. |
| format | Article |
| id | doaj-art-c58b540cd7ae4686bf7133a1d995e924 |
| institution | Kabale University |
| issn | 2169-3536 |
| language | English |
| publishDate | 2025-01-01 |
| publisher | IEEE |
| record_format | Article |
| series | IEEE Access |
| spelling | doaj-art-c58b540cd7ae4686bf7133a1d995e9242025-08-20T03:47:28ZengIEEEIEEE Access2169-35362025-01-0113862038621710.1109/ACCESS.2025.356991311003953Octagram Propagation Matching for Multi-Scale View Stereopsis (OPM-MVS)Ray L. Khuboni0https://orcid.org/0000-0002-9035-6873Hongjun Xu1https://orcid.org/0000-0002-5768-1965School of Engineering, University of KwaZulu-Natal (UKZN), Howard College Campus, Durban, South AfricaSchool of Engineering, University of KwaZulu-Natal (UKZN), Howard College Campus, Durban, South AfricaMultiple view stereopsis (MVS) algorithms have achieved substantial advancements in 3D reconstruction accuracy and completeness. However, significant challenges persist with texture-less regions, occlusions, thin structures, repetitive structures, and non-Lambertian surfaces due to unreliable photometric consistency and insufficient geometric information. To address these challenges, we propose octagram propagation matching for multi-scale view stereopsis (OPM-MVS), a novel framework for efficient and accurate depth map estimation. The framework introduces octagram checkerboard propagation as a structural sampling method that enhances hypotheses propagation by aggregating cost information from overlapping neighbouring regions along structural lines. This approach ensures accurate and consistent hypotheses selection while improving spatial propagation efficiency. To refine depth estimation, we incorporate multi-scale depth confidence-guided geometric consistency, propagating reliable depth estimates from coarse to finer scales and minimizing propagation error. A dual bilateral weighted normalized cross-correlation is incorporated to reduce irrelevant pixel influence and improve correspondence in texture-less and repetitive regions. A multi-hypotheses joint view selection strategy is guided by neighbouring joint view probability to aggregate the subset views for cost matching and integrates effectively with the proposed propagation method. Experimental evaluations on public benchmarks includes ETH3D and Tanks and Temples to demonstrate the effectiveness of OPM-MVS in achieving superior completeness while maintaining comparable accuracy to COLMAP, ACMH, ACMM and other surveyed MVSNet methods. The framework delivers state-of-the-art performance by accurately reconstructing depth in challenging regions resulting in complete and reliable 3D models. Notably, OPM-MVS achieves an F1-score of 83.50% on the overall ETH3D test dataset under a 2 cm threshold which outperforms COLMAP (73.01%), ACMH (75.89%), ACMM (80.78%), DP-MVS (83.11%), MVP-Stereo (76.12%), TransMVSFormer (75.39%), TAPA-MVS (79.15%), and UGNet (80.83%). These results establish OPM-MVS as a promising solution for high-resolution and large-scale multi-view stereo applications.https://ieeexplore.ieee.org/document/11003953/Checkerboard propagationmulti-scale patchmatchmulti-view stereostructured region informationconfidence-guided geometric consistencytexture-less regions |
| spellingShingle | Ray L. Khuboni Hongjun Xu Octagram Propagation Matching for Multi-Scale View Stereopsis (OPM-MVS) IEEE Access Checkerboard propagation multi-scale patchmatch multi-view stereo structured region information confidence-guided geometric consistency texture-less regions |
| title | Octagram Propagation Matching for Multi-Scale View Stereopsis (OPM-MVS) |
| title_full | Octagram Propagation Matching for Multi-Scale View Stereopsis (OPM-MVS) |
| title_fullStr | Octagram Propagation Matching for Multi-Scale View Stereopsis (OPM-MVS) |
| title_full_unstemmed | Octagram Propagation Matching for Multi-Scale View Stereopsis (OPM-MVS) |
| title_short | Octagram Propagation Matching for Multi-Scale View Stereopsis (OPM-MVS) |
| title_sort | octagram propagation matching for multi scale view stereopsis opm mvs |
| topic | Checkerboard propagation multi-scale patchmatch multi-view stereo structured region information confidence-guided geometric consistency texture-less regions |
| url | https://ieeexplore.ieee.org/document/11003953/ |
| work_keys_str_mv | AT raylkhuboni octagrampropagationmatchingformultiscaleviewstereopsisopmmvs AT hongjunxu octagrampropagationmatchingformultiscaleviewstereopsisopmmvs |